yanknn

Yanknn is a lightweight open-source neural network framework designed for education and rapid experimentation. The project emphasizes clarity and simplicity, offering a minimal set of building blocks that are easy to inspect and extend. It supports common supervised learning workflows and is designed to run on standard CPU environments without requiring specialized hardware.

The framework provides automatic differentiation through a reverse-mode computational graph, enabling gradient computation for arbitrary sequences of tensor operations. Its API centers on composable layers (dense, convolutional, recurrent), activation functions, and loss terms, with a straightforward training loop that users can customize. Yanknn aims to balance readability with reasonable performance, implementing essential optimizations in compiled components while keeping the Python interface approachable.

Core components include a Tensor type wrapping multi-dimensional arrays, a lightweight autograd engine, a module and layer abstraction, and an optimizer library with SGD, Adam, and RMSProp variants. The data pipeline includes simple dataset utilities and data loaders, supporting basic batching, shuffling, and pre-processing helpers. While not as feature-rich as larger frameworks, yanknn frequently serves as an educational stepping stone to understand how neural networks are built under the hood.

Development and licensing: yanknn is released under an open-source license and maintained by a community of contributors. Documentation emphasizes tutorials and examples that illustrate common patterns such as linear models, multilayer perceptrons, and small convolutional networks. The project encourages educational use and modular extension, with plans to expand backends and improve GPU support in future releases.